Every AI is adding memory. None of it is yours.
Your assistants remember you now, so that's not the problem anymore. The problem is whose memory it is. Each one keeps its own private record of you, inside its own walls, knowing only the slice of you that passed through that app. You can't carry it to the model you'll run next month, hand the relevant parts to the people you work with, or set who sees what. Common is the other way around: the memory is yours, any model or harness reads and writes to it, and you decide who else can. Own your memory. Rent the intelligence.
Every assistant ships memory now. They all keep it the same way.
ChatGPT remembers you, Claude remembers you, Mem0 and the rest sell it as a service. Every one of them holds that memory inside its own product, tied to one account, knowing only what happened in its own window. Run more than one model or harness, the way most of us already do, and you don't have one memory of your work. You have several partial ones that never talk to each other, that you can't move, and that nobody else can read. The memory exists. It's just the wrong shape, and it isn't yours.
"But my AI already connects to all my tools."
Connectors let a model read your stuff at the moment you ask: it opens the doc, runs the query, and then the turn ends. Nothing it learns or produces is kept, and nothing crosses to another model, another person, or tomorrow's session. That's read access, not memory. Wiring an assistant into your tools makes it well-informed for one conversation; it doesn't leave you holding anything afterward.
A memory layer that lives outside the model.
One place that holds the context your work runs on, kept in a form you own. Any assistant you use plugs into it through the tools it already speaks, reads what it needs, and writes back what it worked out. The model stops being where your memory lives and becomes what it always was: the thing currently reading from it. Point a different model at it tomorrow and that one picks up where the last left off.
You change models constantly. Your memory shouldn't notice.
Nobody stays on one model. You move for price, for latency, for a longer context window, for a capability that shipped last week, for a harness that just feels better to work in. The model and the harness are rentals you rotate through, and you should rotate freely. What you can't afford to rebuild on every switch is the context underneath, and today that's exactly what gets stranded with the vendor you're leaving. Common keeps it with you, so swapping the intelligence costs you nothing.
| Memory space | You | Spouse | Cofounder | Accountant |
|---|---|---|---|---|
| Household | ● | ● | — | — |
| Acme, the company | ● | — | ● | — |
| Acme · the books | ● | — | ● | ● |
| Deal: Northwind | ● | — | ● | — |
Why this is the layer worth owning.
Every platform shift hands the durable value to whoever owns the part that persists underneath it. The models are commoditizing fast and interchangeable by design; that's the point of renting them. What compounds and stays scarce is the accumulated, permissioned context they all draw on, and right now no one owns it. That's the layer Common is.
the internet itself
a computer in every pocket
intelligence on tap
| What exists | What it is | What Common does instead |
|---|---|---|
| ChatGPT / Claude memory | The assistant keeps a private record of you, inside the vendor's product, for one account. | A memory you hold, not the vendor: portable to any model, shareable, with a record of who changed what. |
| Agent-memory APIsMem0 · Zep · Letta · Supermemory | A hosted store one app reads and writes inside its own loop. | Not wired into one app's loop — an owned, portable layer shared across people, projects, and the agents they each run. |
| claude-mem & CLAUDE.md | Local notes and compressed sessions one harness reads at startup. | A synced, multiplayer memory any harness can read, with per-project access control and history. |
| Cloudflare Agent Memory | A managed memory store bound to one platform's agents. | Works with whatever you run, and it's yours — no platform lock-in. |
| Dropbox · Drive · iCloud | Generic file sync — your files mirrored across devices. | Built for agents: fills with what you reference and what they produce, with fine-grained access for people and agents alike. |
How it works, in a minute
Common is a synced store any model plugs into through the tools it already speaks. It fills two ways: what you point an agent at gets mirrored in, and what your agents produce gets written back. Every change is attributed and reversible, so you can see which person and which agent did what, and roll it back. It stays in sync across your machines and the people you've shared it with.
The intelligence gets cheaper every quarter. Your memory shouldn't reset every time you switch.
You'll keep renting whatever model is best this month, and you should. The part that lasts is the context you and your agents have built up. Own that, share it on your terms, and carry it across every model you'll ever run.
Own your memory.
Rent the intelligence.